Removing common artefacts in high-resolution diffusion-weighted MRI using novel processing tools

A common oscillation artefact in high-resolution magnetic resonance imaging (MRI), the so-called Gibbs ringing, causes physically implausible signals (PIS) in diffusion-weighted imaging (DWI). Despite omnipresence of this artefact in most MRI techniques, standard neuroimaging preprocessing such as the state-of-the-art pipeline by FSL fail to address this issue. However, recently proposed artefact correction methods (Kellner et al., 2016, and Siemens Healthcare GmbH, Erlangen) have not yet been evaluated systematically. Therefore, we aimed to evaluate these tools applied on DWI data in a reasonably large sample. We collected DWI scans of 121 healthy participants (60f, 19-54 years). Raw images were denoised (MRTRIX) and then processed according to the standard FSL pipeline, i.e. brain mask estimation, eddy-current and motion correction and tensor fitting. We applied additionally either the “unring” tool (Kellner et al.) after denoising or the on-line Siemens filter on the raw images. Preliminary visual inspection revealed consistently sharper images when comparing “unringed” scans to on-line filtered or standardly processed scans. To quantitatively evaluate those visually evident differences, we are currently investigating image quality improvements by the applied artefact removal techniques regarding data accuracy and occurrence of PIS. Further, we examine effects by the choice of removal technique on postprocessing statistics. We report qualitative improvements of DWI images by applying the novel “unring” tool on a large sample size. This points to the need and feasibility that the common oscillation artefact should be taken into account using the “unring” tool as a standard step in processing high-resolution DWI images